Research On Intelligent Auxiliary Diagnosis Technology Of Coronary Artery Blockage Driven By Deep Learning

As one of the most common and harmful cardiovascular diseases in clinical practice,coronary heart disease(CHD)is usually caused by coronary artery occlusion,so it is particularly important to carry out early diagnosis and intervention.Phonocardiogram(PCG)signal is an important physiological signal reflecting human heart activity,which is collected by cardiac auscultation and can objectively reflect the state of heart and cardiovascular system.Coronary artery occlusion will cause turbulence in coronary blood,which will result in high-frequency murmur in diastolic interval of PCG.Before coronary artery occlusion develops to abnormal electrocardiogram,high-frequency murmur and distortion in PCG contain important information for diagnosis,and the analysis of PCG signals has important application value for intelligent auxiliary diagnosis of coronary artery occlusion.In this thesis,the intelligent auxiliary diagnosis technology of coronary artery occlusion driven by deep learning is studied,and location,smallsample generation and coronary artery occlusion recognition of PCG diastolic phase are studied.On the basis of baseline calibration and wavelet denoising preprocessing,a PCG segmentation algorithm based on hidden Markov cardiac cycle is proposed to realize accurate location of PCG diastolic phase;an improved denoising diffusion probability model is designed to solve the problem of insufficient samples during PCG diastolic phase;Convolutional Neural Networks(CNN)model based on Alexnet is constructed to realize the effective identification of coronary artery occlusion.The main work of this thesis is as follows:(1)A PCG signal acquisition system is developed,which is composed of PCG amplification,battery management,transducer,wireless transmission and OLED display modules.It can dynamically collect patient PCG data from four positions,such as aortic valve area,and carry out noise elimination processing,so as to realize high-quality PCG signal acquisition and real-time transmission acquisition.(2)A PCG segmentation algorithm based on Hidden Markov Cardiac Cycle(HMCC)is proposed.Baseline calibration and wavelet denoising are used to realize signal standardization;Hilbert extraction envelope combined with cardiac cycle are used to accurately locate the corresponding relationship between peak value and S1 and S2.The improved hidden Markov algorithm is used to calculate the initial state distribution of PCG,and the duration of PCG interval is calculated with Viterbi algorithm,so as to realize the accurate segmentation and location of PCG diastolic phase.(3)Aiming at the problem of insufficient PCG diastolic signal samples,a PCG sample generation algorithm based on Denoising Diffusion Probabilistic Models(DDPM)is designed.On the basis of diffusion model combined with Unet noise extraction network structure and parameters are adjusted according to the experimental environment and the number of samples to achieve high-quality PCG diastolic signal generation.(4)In order to effectively identify coronary occlusion symptoms,a diagnostic model of coronary occlusion based on Alexnet is constructed.One-dimensional PCG signals are visualized by Gramian angular difference field(GADF)as the input of convolution neural network.Transfer learning technology is adopted to optimize network parameters of Alexnet by fine-tune.5238 diastolic PCG segmented by DLUTHSDB database and 784 diastolic PCG generated by actual collection and DDPM are selected to evaluate the performance of the model,and the accuracy of the model reached 98.92% and 98.59%,respectively.In this thesis,the intelligent auxiliary diagnosis technology of coronary artery occlusion driven by deep learning is proposed,which can improve the accuracy of coronary occlusion diagnosis to a certain extent,realize effective auxiliary diagnosis of coronary occlusion,and provide new ideas for intelligent diagnosis of PCG,with good application prospects.